Training device, inference device, training inference device, program, training inference system, training method, and inference method

Abstract

In the present invention, a pattern detection device (100) comprises: a training data generation unit (101) that generates training data in which a plurality of feature vectors corresponding respectively to a plurality of training target images are arranged in time series; a true value data construction unit (104) that constructs true value data in which a plurality of true value vectors, each indicating true / false for each of the plurality of training target images, are arranged so as to correspond to the time series of the plurality of feature vectors; and a reservoir circuit training unit (107) that uses the training data and the true value data to train a three-layer neural network using reservoir computing.

Description

Learning device, inference device, learning and inference device, program, learning and inference system, learning method and inference method

[0001] The present disclosure relates to a learning device, an inference device, a learning inference device, a program, a learning inference system, a learning method, and an inference method.

[0002] Machine learning is divided into convolutional neural networks (CNNs) and recurrent neural networks (RNNs). CNNs are a deep learning technology that has made progress in improving the accuracy of static pattern detection. On the other hand, RNNs have recurrent connections that CNNs do not have, and therefore have been used exclusively for time-series signal learning.

[0003] Generally, CNNs require larger hardware resources than RNNs, whereas reservoir computing, a technique for RNNs, has the advantage of being able to train with fewer resources than other CNN and RNN techniques. Reservoir computing also has the advantage that it does not require training inside the hidden layer, allowing the hidden layer to be implemented as a circuit using a general physical medium.

[0004] Although reservoir computing has been widely applied to time-series signal learning, its application range can be expanded by applying it to the learning and detection of static two-dimensional patterns. As an example of such technology, a method for implementing two-dimensional pattern detection learning using reservoir computing is shown in Non-Patent Document 1.

[0005] Paugam-Moisy et al. , “Delay learning and polychronization for reservoir computing”, Neurocomputing 71(7-9), pp. 1143-1158, 2008.

[0006] In conventional technologies, it is necessary to learn the coupling constants between the hidden layer units of reservoir computing, which means that the original feature of reservoir computing, that is, the ability to use ordinary physical media to construct the hidden layer, is lost. Therefore, even if there is a physical medium that can be used to construct the hidden layer, there is a problem in that it becomes more difficult to construct a learning network using it.

[0007] Therefore, one or more aspects of the present disclosure aim to make it possible to easily construct a learning network using reservoir computing.

[0008] A learning device according to one aspect of the present disclosure is characterized by comprising: a learning data generation unit that generates learning data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in a time series; a true value data construction unit that constructs true value data in which a plurality of true value vectors indicating the truth or falsity of each of the plurality of training target images are arranged so as to correspond to the time series of the plurality of feature vectors; and a learning unit that uses the learning data and the true value data to train a three-layer neural network using reservoir computing.

[0009] An inference device according to one aspect of the present disclosure is characterized by comprising: an application data generation unit that generates a feature vector as application data from an inference target image, which is an image that is the subject of inference; training data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in chronological order; and true value data in which a plurality of true value vectors indicating the authenticity of each of the plurality of training target images are arranged so as to correspond to the chronological order of the plurality of feature vectors, to train a three-layer neural network using reservoir computing; generating a trained three-layer neural network using coupling constants of the three-layer neural network obtained as a result of the training; and inputting the application data into the trained three-layer neural network to infer the authenticity of the inference target image.

[0010] A learning and inference device according to one aspect of the present disclosure is characterized by comprising: a training data generation unit that generates training data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in time series; a true value data construction unit that constructs true value data in which a plurality of true value vectors indicating the authenticity of each of the plurality of training target images are arranged so as to correspond to the time series of the plurality of feature vectors; a learning unit that uses the training data and the true value data to train a three-layer neural network using reservoir computing and acquires coupling constants of the three-layer neural network as a result of the training; an application data generation unit that generates feature vectors as application data from an inference target image, which is an image that is the subject of inference; and a processing unit that uses the acquired coupling constants to generate a trained three-layer neural network and infers the authenticity of the inference target image by inputting the application data to the trained three-layer neural network.

[0011] A program according to a first aspect of the present disclosure causes a computer to function as: a training data generation unit that generates training data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in time series; a true value data construction unit that constructs true value data in which a plurality of true value vectors indicating the truth or falsity of each of the plurality of training target images are arranged so as to correspond to the time series of the plurality of feature vectors; and a learning unit that uses the training data and the true value data to train a three-layer neural network using reservoir computing.

[0012] A program according to a second aspect of the present disclosure causes a computer to function as an application data generation unit that generates a feature vector as application data from an inference target image, which is an image that is the subject of inference, and a processing unit that uses training data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in chronological order and true value data in which a plurality of true value vectors indicating the authenticity of each of the plurality of training target images are arranged so as to correspond to the chronological order of the plurality of feature vectors, trains a three-layer neural network using reservoir computing, generates a trained three-layer neural network using coupling constants of the three-layer neural network obtained as a result of the training, and inputs the application data into the trained three-layer neural network to infer the authenticity of the inference target image.

[0013] A learning inference system according to one aspect of the present disclosure is characterized by comprising: a training data generation unit that generates training data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in time series; a true value data construction unit that constructs true value data in which a plurality of true value vectors indicating the authenticity of each of the plurality of training target images are arranged so as to correspond to the time series of the plurality of feature vectors; a learning unit that uses the training data and the true value data to train a three-layer neural network using reservoir computing and acquires coupling constants of the three-layer neural network as a result of the training; an application data generation unit that generates feature vectors as application data from an inference target image, which is an image that is the subject of inference; and a processing unit that uses the acquired coupling constants to generate a trained three-layer neural network and infers the authenticity of the inference target image by inputting the application data to the trained three-layer neural network.

[0014] A learning method according to one aspect of the present disclosure is characterized in that it generates learning data in which a plurality of feature vectors corresponding to a plurality of training target images are arranged in a time series, constructs true-value data in which a plurality of true-value vectors indicating the truth or falsity of each of the plurality of training target images are arranged to correspond to the time series of the plurality of feature vectors, and uses the learning data and the true-value data to train a three-layer neural network using reservoir computing.

[0015] An inference method according to one aspect of the present disclosure is characterized in that it generates a feature vector as application data from an inference target image, which is an image to be inferred, and uses training data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in chronological order, and true value data in which a plurality of true value vectors indicating the authenticity of each of the plurality of training target images are arranged so as to correspond to the chronological order of the plurality of feature vectors, to train a three-layer neural network using reservoir computing, and generates a trained three-layer neural network using the coupling constants of the three-layer neural network obtained as a result of the training, and infers the authenticity of the inference target image by inputting the application data into the trained three-layer neural network.

[0016] According to one or more aspects of the present disclosure, a learning network using reservoir computing can be easily constructed.

[0017] 1 is a block diagram schematically showing the configuration of a pattern detection device as a learning and inference device according to a first embodiment. FIG. 1 is a schematic diagram showing an example of a training image, which is a sample image including a target for pattern detection. FIG. 2 is a schematic diagram showing an example of extracting, from a training image, a plurality of partial images for determining whether an image is a genuine image or a fake image. FIG. 3 is a schematic diagram for explaining an example of a training image set. FIG. 4 is a schematic diagram for explaining processing in a training phase according to the first embodiment. FIG. 5 is a schematic diagram for explaining an example of learning a plurality of training images according to the first embodiment. FIG. 6 is a schematic diagram for explaining processing in an application phase according to the first embodiment. FIG. 7 is a block diagram schematically showing the configuration of a computer. FIG. 8 is a block diagram schematically showing the configuration of a pattern detection device as a learning and inference device according to a second embodiment. FIG. 9 is a schematic diagram for explaining processing in a training input data designation unit and an orthogonal filter application unit in a training phase according to the second embodiment. FIG. 10 is a schematic diagram showing an example of a two-dimensional Walsh-Hadamard filter with a rank N=8. FIG. 11 is a schematic diagram for explaining processing in an application phase according to the second embodiment. FIG. 12 is a block diagram schematically showing the configuration of a pattern detection device according to a third embodiment. 10 is a schematic diagram for explaining the processing in the learning input data specifying unit, the orthogonal filter applying unit, and the range converting unit in the learning phase in embodiment 3. FIG. 11 is a graph showing a hyperbolic tangent function. FIG. 12 is a schematic diagram for explaining the processing in the application phase in embodiment 3.

[0018] 1 is a block diagram showing a schematic functional configuration of a pattern detection device 100 as a learning and inference device according to embodiment 1. The pattern detection device 100 includes a learning data generation unit 101, a true value data construction unit 104, a reservoir circuit learning unit 107, an application data input unit 111, a reservoir circuit processing unit 112, and an application data output unit 113.

[0019] The pattern detection device 100 operates in two phases, a learning phase and an application phase, in a processing process that uses an Echo State Network, which is a discrete-time implementation of reservoir computing.

[0020] The training data generation unit 101 generates training data in which a plurality of feature vectors corresponding to a plurality of training images are arranged in chronological order. In the first embodiment, the training data generation unit 101 extracts pixel values ​​in a predetermined order from each of the plurality of training images, which are two-dimensional images, and generates each of the plurality of feature vectors by arranging the extracted pixel values ​​in the extraction order. The training data generation unit 101 includes a training input data specification unit 102 and a training feature vector storage unit 103.

[0021] In the learning phase, the learning input data designation unit 102 transforms one learning image to generate a feature vector. Note that individual partial images included in the learning image are also referred to as learning target images. The learning input data designation unit 102 transforms multiple learning target images to generate multiple feature vectors. Then, in the learning phase, the learning input data designation unit 102 configures these multiple feature vectors to have a fixed time length, thereby generating learning feature vectors as learning data.

[0022] The training feature vector storage unit 103 stores the training feature vectors generated by the training input data designation unit 102 in the training phase.

[0023] The true-value data constructing unit 104 constructs true-value data in which a plurality of true-value vectors indicating the trueness or falseness of each of a plurality of training images are arranged to correspond to a time series of a plurality of feature vectors generated from the plurality of training images. The true-value data constructing unit 104 includes a training output data specifying unit 105 and a training true-value vector storage unit 106.

[0024] In the learning phase, the learning output data specification unit 105 generates a true value vector indicating true value data corresponding to each feature vector included in the learning feature vector, as a learning true value vector, so that the true value vector corresponds one-to-one with the true value of each feature vector included in the learning feature vector generated by the learning input data specification unit 102.

[0025] In the learning phase, the training true value vector storage unit 106 stores the training true value vectors generated by the training output data designation unit 105 so that each true value vector included in the training true value vector corresponds to each feature vector included in the training feature vector stored in the training feature vector storage unit 103. Then, the training true value vector storage unit 106 inputs the training true value vectors as true value data to the reservoir circuit learning unit 107.

[0026] The reservoir circuit learning unit 107 functions as a learning unit that uses the learning data generated by the learning data generation unit 101 and the true value data constructed by the true value data construction unit 104 to train a three-layer neural network using reservoir computing.

[0027] In the learning phase, the reservoir circuit learning unit 107 applies the Echo State Network learning algorithm to the three-layer neural network based on the learning feature vectors stored in the learning feature vector storage unit 103 and the learning true-value vectors stored in the learning true-value vector storage unit 106. Then, the reservoir circuit learning unit 107 stores the coupling constants of the three-layer neural network as the learning results of the Echo State Network learning algorithm.

[0028] The application data input unit 111 in the first embodiment functions as an application data generation unit that generates, as application data, a feature vector from an inference target image, which is an image to be inferred. In the first embodiment, the application data input unit 111 extracts pixel values ​​in a predetermined order from the inference target image, which is a two-dimensional image, and generates application data by arranging the extracted pixel values ​​in the extraction order.

[0029] For example, in the application phase, the application data input unit 111 receives a new input of a two-dimensional image and converts the two-dimensional image to obtain its feature vector. This two-dimensional image becomes an inference target image.

[0030] The reservoir circuit processing unit 112 generates a trained three-layer neural network using the coupling constants of the three-layer neural network obtained as a result of learning in the reservoir circuit learning unit 107, and functions as a processing unit that infers the authenticity of the image to be inferred by inputting application data from the application data input unit 111 into the trained three-layer neural network.

[0031] For example, in the application phase, the reservoir circuit processing unit 112 constructs a trained three-layer neural network using the coupling constants stored in the reservoir circuit learning unit 107. The reservoir circuit processing unit 112 applies a trained Echo State Network to the trained three-layer network. As shown in FIG. 7 , which will be described later, the trained Echo State Network includes an input layer 112a, an intermediate layer 112b, and an output layer 112c. The reservoir circuit processing unit 112 then inputs the feature vector acquired by the application data input unit 111 to the input layer 112a as input data.

[0032] The application data output unit 113 acquires the output data from the output layer 112c and outputs it as a determination result.

[0033] In this embodiment, a two-dimensional image is used as an example of a pattern detection target. A plurality of partial images extracted from a learning image are set as a learning input image set. Each of the plurality of partial images is a learning target image. Of the plurality of partial images, a partial image that includes the detection target is set as a true image (true), and other partial images are set as false images (false). When a partial image is a true image, a true vector (1, 0) Ｔ (T is the transpose of the vector) If the partial image is a false image, the true vector (0, 1) Ｔ is set as true value data for two numbers of output layer units, and is set as a training output data set associated with a plurality of partial images included in the training input image set. Here, an example of operation when a training image set consisting of a training input image set and a training output data set is trained using an Echo State Network will be described, divided into a training phase and an application phase.

[0034] First, the learning phase will be described. Fig. 2 is a schematic diagram showing an example of a learning image, which is a sample image containing a pattern detection target. Here, six apples present in the learning image are used as an example of the detection target.

[0035] For this reason, for example, as shown in FIG. 3, in order to determine whether an image is genuine or fake from the learning image IM, multiple rectangular frames are specified, and the images within those multiple rectangular frames are extracted as partial images P01 to P12.

[0036] 4 is a schematic diagram illustrating an example of a training image set. In the training image set, true images that are the detection target are indicated as "true," and false images that are not the detection target are indicated as "false."

[0037] The normal image shown in Fig. 4 is a partial image extracted from a rectangular frame containing six apples present in the learning image IM shown in Fig. 3. In order to learn this normal image as a detection target, a true vector (1, 0) Ｔ is given as the true data.

[0038] The fake image shown in Fig. 4 is a partial image extracted from a randomly designated rectangular frame in an area that does not include six apples present in the learning image IM shown in Fig. 3. In order to learn this fake image as a detection target, a true value vector (0, 1) Ｔ is given as the true data.

[0039] Here, a process for learning one partial image having a true image or a false image as a true value from a learning image set using an Echo State Network will be described with reference to FIG.

[0040] 5 is a schematic diagram for explaining the processing in the learning phase. First, two-dimensional image data representing one partial image used for learning the Echo State Network is stored in a learning image input buffer 102 a included in the learning input data designation unit 102.

[0041] Next, in order to convert the partial image represented by the two-dimensional image data stored in the training image input buffer 102a into a feature vector, the training input data designation unit 102 converts the resolution of the partial image and stores the converted data in the feature vector conversion image buffer 102b.

[0042] Next, the learning input data designation unit 102 converts the resolution-converted converted partial image into a feature vector and stores the vector in the feature vector buffer 102 c. To vectorize this converted partial image, the learning input data designation unit 102 may, for example, start from the upper left corner pixel of the converted partial image and read pixel values ​​one pixel at a time in the leftward direction, proceeding from the top row to the bottom row, and continue reading pixel values ​​down to the bottom right pixel.

[0043] Next, the true value of one partial image input to the learning input data specification unit 102, in other words, the true value vector associated with a true image if the image contains the detection target, or a false image if not, is stored in the true value vector storage buffer 105a of the learning output data specification unit 105.

[0044] Here, the number of elements of the true vector may be set to a value equal to or greater than 1. For example, if the number of elements of the true vector is set to 2, the normal image will have a column vector (1, 0) Ｔ , and the fake image has a column vector (0,1) Ｔ The learning can be performed by associating the above.

[0045] As described above, by performing appropriate network initial settings for the feature vectors obtained from the partial images stored in the feature vector buffer 102c, the true value vectors stored in the true value vector storage buffer 105a, and the Echo State Network of the reservoir circuit learning unit 107, it is possible to learn two images, a true image and a false image, in the Echo State Network using the learning algorithm described in Reference 1 below.

[0046] Here, the learning of the Echo State Network requires learning only the coupling constants from the intermediate layer to the output layer, and a specific calculation method for this will be described. First, the reservoir circuit learning unit 107 calculates the coupling constant matrix from the input layer 107a to the intermediate layer 107b for the connections that mutually connect the input layer 107a, the intermediate layer 107b, and the output layer 107c of the Echo State Network by W ｉｎ , the mutual coupling constant matrix of the internal units of the hidden layer 107b is W ｒｅｓ , the coupling constant matrix from the output layer 107c to the intermediate layer 107b is W ｆｂ Furthermore, when the input signal to the input layer 107a at discrete time k is u[k], the state of the internal unit of the hidden layer 107b is x[k], and the output signal of the output layer 107c is y[k], the relationship of the following equation (1) holds.

[0047]

[0048] In equation (1), λ is called the leaking rate and indicates the extent to which the activity of x[k] from one time point before is reflected in the activity of x[k+1], in other words, the extent to which the past internal state history of the hidden layer is dragged along the time axis. When λ = 1, x[k] is not reflected at all in the update of x[k+1], in other words, the internal state history is not dragged along the time axis at all.

[0049] Furthermore, the coupling constant matrix from the intermediate layer 107b to the output layer 107c is W ｏｕｔ , u[k] and x[k] from time 1 to T are grouped as column vectors, and X is a matrix, and y[k], which is true value data, is grouped as column vectors. ｔａｒｇｅｔ When X and Y are ｔａｒｇｅｔ There is a relationship between them expressed by the following formula (2).

[0050]

[0051] Using the Ridge Regression described in the above-mentioned Reference 1, we can convert equation (1) into W ｏｕｔ When solving for this, the following equation (3) is obtained.

[0052]

[0053] As a result, the feature vector derived from the partial image, X consisting of the internal state of the intermediate layer 107b, and Y consisting of the true value vector are obtained. ｔａｒｇｅｔ So, W ｏｕｔ can be calculated.

[0054] Note that equation (3) shows the calculation process of ridge regression, and β is an optimization parameter in ridge regression. If ridge regression is not used for equation (2) but the Moore-Penrose pseudo-inverse matrix is ​​used to calculate W ｏｕｔ When directly calculating, the solution is obtained by the following equation (4).

[0055]

[0056] However, since the amount of calculation required for the pseudo-inverse matrix in equation (4) increases when the dimension of X is large, Ridge Regression is used to avoid the pseudo-inverse matrix calculation and reduce the amount of calculation. In this case, the optimization calculation using Ridge Regression is given by the following equation (5).

[0057]

[0058] In equation (5), if β||W ｏｕｔ || ２ If there is no term, W ｏｕｔ If the norm size of becomes large, the output of the Echo State Network becomes unstable. Therefore, in the calculation process of equation (5), β||W ｏｕｔ || ２ Optimization calculation is performed including the term, and the norm size is adjusted to W ｏｕｔ get.

[0059] An example of having the Echo State Network learn one partial image has been described above. Here, an example of having the Echo State Network learn multiple learning images, each assigned a true value for a positive image or a false image, will be described with reference to FIG. 6 .

[0060] 6, the feature vectors converted by the learning input data designation unit 102 are used as the learning images, which are multiple two-dimensional images to be learned by the Echo State Network. The process of converting each two-dimensional image into a feature vector may be the same as the procedure described above with reference to FIG.

[0061] In FIG. 6, the learning input data designation unit 102 designates n (n is an integer of 2 or more) normal images T １ , T ２ , ..., T ｎ and m (m is an integer equal to or greater than 2) false images F １ , F ２ , ..., F ｍ are stored in chronological order in the form of feature vectors. Furthermore, for each of the true and false images, l consecutive feature vectors are stored.

[0062] In other words, the learning input data designation unit 102 stores a total of l×(n+m) feature vectors in chronological order, which constitute the column vectors of the matrix X in the above formula (2). Corresponding to this chronological order, the learning output data designation unit 105 stores a total of l×(n+m) true value vectors associated with each of the true and false images in chronological order, which constitute the column vectors of the matrix Y in the above formula (2). ｔａｒｇｅｔ It forms a column vector of

[0063] As a specific true vector, as described above, for the normal image, the column vector (1, 0) Ｔ , the fake image has a column vector (0,1) Ｔ It is sufficient if the following can be associated.

[0064] Next, the application phase in the first embodiment will be described with reference to Fig. 7. First, two-dimensional image data representing one partial image is stored in the application image input buffer 111a of the application data input unit 111.

[0065] Next, in order to convert the partial image represented by the two-dimensional image data stored in the application image input buffer 111a into a feature vector, the application data input unit 111 converts the resolution of the partial image and stores the converted data in the image buffer for feature vector conversion 111b.

[0066] Next, the application data input unit 111 converts the resolution-converted converted partial image into a feature vector and stores the vector in the feature vector buffer 111 c. To vectorize this converted partial image, the application data input unit 111 may, for example, start from the upper left corner pixel of the converted partial image and read pixel values ​​one pixel at a time in the leftward direction, proceeding from the top row to the bottom row, and continue reading pixel values ​​down to the bottom right pixel, thereby vectorizing the converted partial image.

[0067] The reservoir circuit processing unit 112 generates a trained three-layer neural network using the coupling constants of the three-layer neural network obtained as a result of training in the reservoir circuit training unit 107. The reservoir circuit processing unit 112 applies a trained Echo State Network to the trained three-layer neural network. The reservoir circuit processing unit 112, which has the trained Echo State Network, accepts input of a feature vector constructed by the application data input unit 111 using a new two-dimensional image as a column vector. The trained Echo State Network includes an input layer 112a, an intermediate layer 112b, and an output layer 112c. The reservoir circuit processing unit 112 then calculates an output vector using the trained Echo State Network and stores the output vector in an output vector storage buffer 113a of the application data output unit 113.

[0068] In the application data output unit 113, the determination circuit 113b determines whether the output vector is a true image, meaning that it is a detection target, or a false image, meaning that it is not a detection target. To determine whether the image is a true image or a false image, the determination circuit 113b may, for example, apply threshold value determination to each element of an output vector having two elements, or may perform the determination using linear separation or other machine learning techniques.

[0069] As described above, a learning network using reservoir computing can be easily constructed. Then, by using the Echo State Network, it becomes possible to detect a specific object contained in a static two-dimensional image as a rectangular position in a partial image. To further improve this discrimination ability, a machine learning technique called ensemble learning, described in Reference 2 below, is known, in which high discrimination performance can be ensured by combining multiple discrimination functions with low discrimination performance. Using this ensemble learning, a trained Echo State Network is considered as a single discrimination function, and by combining multiple Echo State Networks, a discrimination function with higher accuracy can be constructed. Reference 2: I. D. Mienye and Y. Sun, “A Survey of Ensemble Learning: Concepts, Algorithms, Applications, and Prospects”, IEEE Access Vol. 10, pp. 99129-99149, 2022.

[0070] In the first embodiment, an input pattern to the reservoir computing circuit is represented as a column vector, each column vector having a fixed time length constituting the input pattern, and a teacher signal corresponding to the classification category of the input pattern is associated with the output data. This allows the input pattern to be associated with a dynamic equilibrium point of the intermediate layer of the reservoir computing circuit during learning by the reservoir computing circuit. This makes it possible to realize pattern detection using the reservoir computing circuit.

[0071] The pattern detection apparatus 100 described above can be realized by, for example, a computer 10 as shown in Fig. 8. The computer 10 includes a storage 11 such as a hard disk drive (HDD) or a solid state drive (SSD), a memory 12, a processor 13 such as a central processing unit (CPU), a processing circuit 14, and a reservoir circuit 15.

[0072] The processing circuit 14 is configured by a single circuit, a composite circuit, a processor operated by a program, a parallel processor operated by a program, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or the like.

[0073] For example, the learning input data designation unit 102, the learning output data designation unit 105, and the application data input unit 111 can be configured by the processing circuit 14. The learning feature vector storage unit 103 and the learning true value vector storage unit 106 can be configured by the memory 12. The reservoir circuit learning unit 107 and the reservoir circuit processing unit 112 can be configured by loading a program stored in the storage 11 into the memory 12, having the processor 13 execute the program, and having the processor 13 utilize the reservoir circuit 15. The application data output unit 113 can be realized by loading a program stored in the storage 11 into the memory 12, and having the processor 13 execute the program.

[0074] The above programs may be downloaded to storage 11 from a recording medium (not shown) via a reader / writer (not shown) or from a network via a communication I / F (Interface) (not shown), and then loaded onto memory 12 and executed by processor 13. Alternatively, the programs may be directly loaded onto memory 12 from a recording medium (not shown) via a reader / writer or from a network via a communication I / F (not shown), and executed by processor 13. In other words, the programs may be provided by a program product such as a recording medium. As described above, pattern detection device 100 can be realized by a processing circuit network.

[0075] 9 is a block diagram showing a schematic configuration of a pattern detection device 200 as a learning and inference device according to embodiment 2. The pattern detection device 200 includes a learning data generation unit 201, a true value data construction unit 104, a reservoir circuit learning unit 107, an application data generation unit 210, a reservoir circuit processing unit 112, and an application data output unit 113. In embodiment 2, an orthogonal filter application unit 220 is provided that functions as part of the learning data generation unit 201 and the application data generation unit 210.

[0076] The true value data construction unit 104, the reservoir circuit learning unit 107, the reservoir circuit processing unit 112, and the application data output unit 113 of the pattern detection device 200 according to the second embodiment are similar to the true value data construction unit 104, the reservoir circuit learning unit 107, the reservoir circuit processing unit 112, and the application data output unit 113 of the pattern detection device 100 according to the first embodiment.

[0077] However, in the application phase, the reservoir circuit processing unit 112 of the pattern detection device 200 according to the second embodiment inputs the feature vectors filtered by the orthogonal filter application unit 220 to the input layer 112a as input data.

[0078] The training data generation unit 201 generates training data in which a plurality of feature vectors corresponding to a plurality of training images are arranged in time series. In the second embodiment, the training data generation unit 201 extracts pixel values ​​in a predetermined order from a processed image that is a result of applying an orthogonal filter to each of the plurality of training images, which are two-dimensional images, and generates each of the plurality of feature vectors by arranging the extracted pixel values ​​in the extraction order.

[0079] The training data generation unit 201 includes a training input data specification unit 202, an orthogonal filter application unit 220, and a training feature vector storage unit 103. The training feature vector storage unit 103 of the pattern detection device 200 according to the second embodiment is similar to the training feature vector storage unit 103 of the pattern detection device 100 according to the first embodiment.

[0080] In the learning phase, the learning input data designation unit 202 provides an image obtained by converting the resolution of one learning image to the orthogonal filter application unit 220 .

[0081] The application data generation unit 210 extracts pixel values ​​in a predetermined order from a processed image that is a result of applying an orthogonal filter to an inference target image, which is a two-dimensional image, and generates application data by arranging the extracted pixel values ​​in the order in which they were extracted. The application data generation unit 210 includes an application data input unit 211 and an orthogonal filter application unit 220.

[0082] In the application phase, the application data input unit 211 accepts a new input of a two-dimensional image, and provides the image obtained by converting the resolution of the two-dimensional image to the orthogonal filter application unit 220 .

[0083] The orthogonal filter application unit 220 generates a feature vector by applying an orthogonal filter to the image from the learning input data designation unit 202. The orthogonal filter application unit 220 then configures the plurality of filtered feature vectors to have a fixed time length, thereby generating a learning feature vector, and provides the feature vector to the learning feature vector storage unit 103. The orthogonal filter application unit 220 also applies an orthogonal filter to the image from the application data input unit 211 to generate a feature vector, and provides the feature vector to the reservoir circuit processing unit 112.

[0084] 10 is a schematic diagram for explaining the processing in the learning input data designation unit 202 and the orthogonal filter application unit 220 in the learning phase in the second embodiment. Here, an example is shown in which a Walsh-Hadamard filter having filter coefficients consisting of two values, "1" and "-1", is used as the orthogonal filter used in the orthogonal filter application unit 220. As in the first embodiment, the processing in the learning phase when learning one image from the learning image set having a true image or a false image as its true value using the Echo State Network is explained, among the operations for converting the input data held in the learning input data designation unit 202.

[0085] First, two-dimensional image data representing one partial image used for learning the Echo State Network is stored in the learning image input buffer 202 a of the learning input data designation unit 202 .

[0086] Next, in order to convert the partial image represented by the two-dimensional image data stored in the training image input buffer 202a into a feature vector, the training input data designation unit 202 converts the resolution of the partial image and stores the converted data in the feature vector conversion image buffer 202b.

[0087] Next, the orthogonal filter application processing unit 220a of the orthogonal filter application section 220 performs filtering on the two-dimensional image stored in the feature vector conversion image buffer 202b, and stores the result in the orthogonal filter output buffer 220b.

[0088] Next, the orthogonal filter application unit 220 generates a feature vector from the data stored in the orthogonal filter output buffer 220b, as in the first embodiment, and stores the feature vector in the feature vector buffer 220c.

[0089] Thereafter, the processing in the learning phase by the Echo State Network is the same as in embodiment 1. Here, an example of an orthogonal filter used in orthogonal filter application processing unit 220a of orthogonal filter application section 220 is the Walsh-Hadamard filter described in Reference 3 below.

[0090] Reference 3: Y. Hel-Or and H. Hel-Or, “Real-Time Pattern Matching Using Projection Kernels”, IEEE Transactions on Pattern Analysis and Machine Intelligence, 27(9), pp. 1430-1445, 2005.

[0091] A Walsh-Hadamard filter can be defined as one-dimensional, and a two-dimensional Walsh-Hadamard filter can be further defined as the product of a one-dimensional and a one-dimensional.

[0092] FIG. 11 shows an example of a two-dimensional Walsh-Hadamard filter with a rank of N=8, in which one element filter consisting of 8×8 coefficients is arranged vertically and horizontally in a manner that refines the spatial frequency, and can be applied to an image of 8×8 pixels vertically and horizontally.

[0093] Here, the rank N of the Walsh-Hadamard filter must be a power of 2, and can be N = 2, 4, 8, 16, 32, .... For each filter shown in Fig. 11, white represents a coefficient of "+1" and black represents a coefficient of "-1." In other words, when applying this Walsh-Hadamard filter to an 8x8 pixel image, no multiplication is required, and filter calculations are performed using only addition and subtraction, enabling high-speed calculation of spatial frequency decomposition.

[0094] Next, the application phase in the second embodiment will be described with reference to Fig. 12. First, two-dimensional image data representing one partial image is stored in the application image input buffer 211a of the application data input unit 211.

[0095] Next, in order to convert the partial image represented by the two-dimensional image data stored in the application image input buffer 211a into a feature vector, the application data input unit 211 converts the resolution of the partial image and stores the converted data in the image buffer for feature vector conversion 211b.

[0096] Next, the orthogonal filter application processing unit 220a of the orthogonal filter application section 220 performs filtering on the two-dimensional image stored in the feature vector conversion image buffer 211b, and stores the result in the orthogonal filter output buffer 220b.

[0097] Next, the orthogonal filter application unit 220 generates a feature vector from the data stored in the orthogonal filter output buffer 220b, as in the first embodiment, and stores the feature vector in the feature vector buffer 220c.

[0098] The reservoir circuit processing unit 112, which has a trained Echo State Network, accepts input of feature vectors constructed by the orthogonal filter application unit 220 using a new two-dimensional image as a column vector. The trained Echo State Network includes an input layer 112a, an intermediate layer 112b, and an output layer 112c. The reservoir circuit processing unit 112 then calculates an output vector using the trained Echo State Network and stores the output vector in an output vector storage buffer 113a of the application data output unit 113.

[0099] In the application data output unit 113, the determination circuit 113b determines whether the output vector is a true image, meaning that it is a detection target, or a false image, meaning that it is not a detection target.

[0100] In the second embodiment, similarly to the first embodiment, in the learning phase, a plurality of learning images may be stored in the learning input data designation unit 202, and a plurality of true value vectors linked to the individual learning images may be stored in the learning output data designation unit 105, and then the reservoir circuit learning unit 107 may learn the Echo State Network.

[0101] In the second embodiment, a Walsh-Hadamard filter is used as the orthogonal filter used in the orthogonal filter application unit 220. However, another orthogonal filter may be used. For example, a two-dimensional image may be converted into a feature vector using a discrete cosine transform (DCT) or a fast Fourier transform (FFT) as introduced in the above-mentioned reference 3.

[0102] In the second embodiment, when the training patterns to be trained by the reservoir computing circuit are mapped into the feature space, the distance within the feature space of training patterns with different true values ​​can be expanded using an orthogonal transformation, which makes it possible to easily construct a training network using reservoir computing and to improve the discrimination performance of pattern detection using the reservoir computing circuit.

[0103] The pattern detection device 200 described above can also be realized by a computer 10 as shown in Fig. 8. For example, the orthogonal filter application unit 220 can also be configured by loading a program stored in the storage 11 into the memory 12 and having the processor 13 execute the program, or by using the processing circuitry 14.

[0104] 13 is a block diagram showing a schematic configuration of a pattern detection device 300 according to embodiment 3. The pattern detection device 300 includes a learning data generation unit 301, a true value data construction unit 104, a reservoir circuit learning unit 107, an application data generation unit 310, a reservoir circuit processing unit 112, and an application data output unit 113. In embodiment 3, an orthogonal filter application unit 320 and a range conversion unit 321 are provided, which function as part of the learning data generation unit 301 and the application data generation unit 310.

[0105] The true value data construction unit 104, reservoir circuit learning unit 107, reservoir circuit processing unit 112, and application data output unit 113 of the pattern detection device 300 according to the third embodiment are similar to the true value data construction unit 104, reservoir circuit learning unit 107, reservoir circuit processing unit 112, and application data output unit 113 of the pattern detection device 100 according to the first embodiment. However, in the application phase, the reservoir circuit processing unit 112 of the pattern detection device 300 according to the third embodiment inputs the feature vector converted in the range conversion unit 321 to the input layer 112a as input data.

[0106] The training data generation unit 301 generates training data in which a plurality of feature vectors corresponding to a plurality of training images are arranged in time series. In the third embodiment, the training data generation unit 301 extracts pixel values ​​in a predetermined order from a transformed image, which is the result of applying a function that transforms pixel values ​​into a predetermined range, to a processed image, which is the result of applying an orthogonal filter to each of the plurality of training images, which are two-dimensional images, and then arranges the extracted pixel values ​​in the order in which they were extracted, thereby generating each of the plurality of feature vectors.

[0107] The training data generation unit 301 includes a training input data specification unit 202, an orthogonal filter application unit 320, a range conversion unit 321, and a training feature vector storage unit 103. The training input data specification unit 202 of the pattern detection device 300 according to the third embodiment is similar to the training input data specification unit 202 of the pattern detection device 200 according to the second embodiment. Furthermore, the training feature vector storage unit 103 of the pattern detection device 300 according to the third embodiment is similar to the training feature vector storage unit 103 of the pattern detection device 300 according to the first embodiment.

[0108] The application data generation unit 310 generates, as application data, a feature vector from an inference target image, which is an image to be inferred. In the third embodiment, the application data generation unit 310 extracts pixel values ​​in a predetermined order from a transformed image, which is a result of applying a function that transforms pixel values ​​into a predetermined value range, to a processed image, which is a result of applying an orthogonal filter to the inference target image, which is a two-dimensional image, and generates application data by arranging the extracted pixel values ​​in the order in which they were extracted.

[0109] The application data generation unit 310 includes an application data input unit 211, an orthogonal filter application unit 320, and a range conversion unit 321. The application data input unit 211 of the pattern detection device 300 according to the third embodiment is similar to the application data input unit 211 of the pattern detection device 200 according to the third embodiment.

[0110] The orthogonal filter application unit 320 applies an orthogonal filter to the image from the learning input data designation unit 202, and provides the filtered image data to the value range conversion unit 321. The orthogonal filter application unit 320 also applies an orthogonal filter to the image from the application data input unit 211, and provides the filtered image data to the value range conversion unit 321.

[0111] In the learning phase, the range conversion unit 321 applies a function that converts a range to image data that has been filtered by the orthogonal filter application unit 320 for an image from the learning input data designation unit 202, thereby generating a feature vector. The range conversion unit 321 then configures multiple feature vectors to which such a function has been applied so that they have a fixed time length, thereby generating a training feature vector, and provides the feature vector to the training feature vector storage unit 103. In the learning phase, the range conversion unit 321 also applies a function that converts a range to image data that has been filtered by the orthogonal filter application unit 320 for an image from the application data input unit 211, thereby generating a feature vector, and provides the feature vector to the reservoir circuit processing unit 112.

[0112] 14 is a schematic diagram for explaining the processing performed by the learning input data designation unit 202, the orthogonal filter application unit 320, and the range conversion unit 321 in the learning phase in the third embodiment. Here, as in the second embodiment, an example is shown in which a Walsh-Hadamard filter having filter coefficients consisting of two values, "1" and "-1", is used as the orthogonal filter used in the orthogonal filter application unit 320. Then, a hyperbolic tangent function (tanh(x)) is used as the nonlinear function for range conversion used in the range conversion unit 321, and an operation is described in which the output value of the Walsh-Hadamard filter output from the orthogonal filter application unit 320 is kept between "-1" and "1" to construct a feature vector.

[0113] First, two-dimensional image data representing one partial image used for learning the Echo State Network is stored in the learning image input buffer 202 a of the learning input data designation unit 202 .

[0114] Next, in order to convert the partial image represented by the two-dimensional image data stored in the training image input buffer 202a into a feature vector, the training input data designation unit 202 converts the resolution of the partial image and stores the converted data in the feature vector conversion image buffer 202b.

[0115] Next, the orthogonal filter application processing unit 320a of the orthogonal filter application section 320 performs filtering on the two-dimensional image stored in the feature vector conversion image buffer 202b, and stores the result in the orthogonal filter output buffer 320b.

[0116] Next, the range conversion unit 321 converts the values ​​of the filtered image data stored in the orthogonal filter output buffer 320b using the range conversion function possessed by the range conversion function application processing unit 321a, and data having the converted values ​​is stored in the range conversion function output buffer 321b.

[0117] Then, the range transformation unit 321 generates a feature vector from the data stored in the range transformation function output buffer 321b, as in the first embodiment, and stores the feature vector in the feature vector buffer 321c.

[0118] An example of the range conversion function that the range conversion function application processing unit 321a has is the hyperbolic tangent function. The hyperbolic tangent function tanh(x) is expressed by the following equation (6), and FIG. 15 shows a graph of this function.

[0119]

[0120] In the second embodiment, a feature vector is constructed directly from each value of the orthogonal filter output buffer 220b, which is the calculation result of the orthogonal filter application unit 220. However, in the third embodiment, even if the distribution of the applied values ​​of the orthogonal filter to the input image varies greatly, the range conversion unit 321 causes tanh(x) to fall within the range of -1 < f(x) < 1, thereby stabilizing the operation of the reservoir circuit learning unit 107 during learning.

[0121] Next, the application phase in the third embodiment will be described with reference to Fig. 16. First, two-dimensional image data representing one partial image is stored in the application image input buffer 211a of the application data input unit 211.

[0122] Next, in order to convert the partial image represented by the two-dimensional image data stored in the application image input buffer 211a into a feature vector, the application data input unit 211 converts the resolution of the partial image and stores the converted data in the image buffer for feature vector conversion 211b.

[0123] Next, the orthogonal filter application processing unit 320a of the orthogonal filter application section 320 performs filtering on the two-dimensional image stored in the feature vector conversion image buffer 211b, and stores the result in the orthogonal filter output buffer 320b.

[0124] Next, the range conversion unit 321 converts the values ​​of the filtered image data stored in the orthogonal filter output buffer 320b using the range conversion function possessed by the range conversion function application processing unit 321a, and data having the converted values ​​is stored in the range conversion function output buffer 321b.

[0125] Then, the range transformation unit 321 generates a feature vector from the data stored in the range transformation function output buffer 321b, as in the first embodiment, and stores the feature vector in the feature vector buffer 321c.

[0126] The reservoir circuit processing unit 112, which has a trained Echo State Network, accepts input of the feature vector constructed by the range conversion unit 321 using a new two-dimensional image as a column vector. The trained Echo State Network includes an input layer 112a, an intermediate layer 112b, and an output layer 112c. The reservoir circuit processing unit 112 then calculates an output vector using the trained Echo State Network and stores the output vector in the output vector storage buffer 113a of the application data output unit 113.

[0127] In the application data output unit 113, the determination circuit 113b determines whether the output vector is a true image, meaning that it is a detection target, or a false image, meaning that it is not a detection target.

[0128] In the third embodiment, similarly to the first embodiment, in the learning phase, a plurality of learning images may be stored in the learning input data designation unit 202, and a plurality of true value vectors linked to the individual learning images may be stored in the learning output data designation unit 105, and then the reservoir circuit learning unit 107 may learn the Echo State Network.

[0129] As described above, according to the third embodiment, before a feature vector is input to the reservoir circuit processing unit 112 having a trained Echo State Network, the range conversion unit 321 can limit the value of each element constituting the feature vector to the range between "-1" and "1." This makes it possible to easily construct a learning network using reservoir computing and stabilize the processing operation of the reservoir circuit processing unit 112.

[0130] In addition, in embodiment 3, by narrowing the range of values ​​of the input data input to the reservoir computing circuit, the operation of the reservoir computing circuit becomes stable, making it possible to improve the identification performance of pattern detection using the reservoir computing circuit.

[0131] In the above-described first to third embodiments, the pattern detection devices 100 to 300 have been described as devices that perform processing in both the learning phase and the inference phase, but the first to third embodiments are not limited to such devices. For example, the first to third embodiments may also be configured as a learning device that performs processing in the learning phase and an inference device that performs processing in the inference phase. Furthermore, the processing performed by the pattern detection devices 100 to 300 according to the first to third embodiments may be distributed among multiple computers that are connected to a network and can send and receive data between them. In other words, the first to third embodiments may be configured as a pattern detection system that is a learning and inference system made up of multiple computers.

[0132] 100, 200, 300 Pattern detection device, 101, 201, 301 Learning data generation unit, 102, 202 Learning input data designation unit, 103 Learning feature vector storage unit, 104 True value data construction unit, 105 Learning output data designation unit, 106 Learning true value vector storage unit, 107 Reservoir circuit learning unit, 210, 310 Application data generation unit, 111, 211 Application data input unit, 112 Reservoir circuit processing unit, 113 Application data output unit, 220, 320 Orthogonal filter application unit, 321 Value range conversion unit.

Claims

1. A learning device comprising: a learning data generation unit that generates learning data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in a time series; a true-value data construction unit that constructs true-value data in which a plurality of true-value vectors indicating the truth or falsity of each of the plurality of training target images are arranged corresponding to the time series of the plurality of feature vectors; and a learning unit that uses the learning data and the true-value data to train a three-layered neural network using reservoir computing.

2. The learning device according to claim 1, wherein each of the plurality of learning target images is a two-dimensional image, and the learning data generation unit generates each of the plurality of feature vectors by extracting pixel values ​​from the two-dimensional image in a predetermined order and arranging the extracted pixel values ​​in the order in which they were extracted.

3. The learning device according to claim 1, wherein each of the plurality of training images is a two-dimensional image, and the training data generation unit extracts pixel values ​​in a predetermined order from a processed image that is a result of applying an orthogonal filter to the two-dimensional image, and generates each of the plurality of feature vectors by arranging the extracted pixel values ​​in the order in which they were extracted.

4. The learning device according to claim 1, wherein each of the plurality of training images is a two-dimensional image, and the training data generation unit extracts pixel values ​​in a predetermined order from a transformed image, which is the result of applying a function that transforms pixel values ​​into a predetermined range, to a processed image, which is the result of applying an orthogonal filter to the two-dimensional image, and generates each of the plurality of feature vectors by arranging the extracted pixel values ​​in the order in which they were extracted.

5. An inference device comprising: an application data generation unit that generates a feature vector as application data from an inference target image, which is an image that is the subject of inference; and a processing unit that uses training data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in chronological order, and true value data in which a plurality of true value vectors indicating the authenticity of each of the plurality of training target images are arranged to correspond to the time series of the plurality of feature vectors, to train a three-layered neural network using reservoir computing, generates a trained three-layered neural network using the coupling constants of the three-layered neural network obtained as a result of the training, and inputs the application data into the trained three-layered neural network to infer the authenticity of the inference target image.

6. The inference device according to claim 5, wherein the image to be inferred is a two-dimensional image, and the application data generation unit generates the application data by extracting pixel values ​​from the two-dimensional image in a predetermined order and arranging the extracted pixel values ​​in the order in which they were extracted.

7. The inference device described in claim 5, characterized in that the image to be inferred is a two-dimensional image, and the application data generation unit generates the application data by extracting pixel values ​​in a predetermined order from a processed image that is the result of applying an orthogonal filter to the two-dimensional image, and arranging the extracted pixel values ​​in the order in which they were extracted.

8. The inference device according to claim 5, wherein the image to be inferred is a two-dimensional image, and the application data generation unit generates the application data by extracting pixel values ​​in a predetermined order from a transformed image, which is the result of applying a function that transforms pixel values ​​into a predetermined value range, to a processed image, which is the result of applying an orthogonal filter to the two-dimensional image, and arranging the extracted pixel values ​​in the order in which they were extracted.

9. A learning and inference device comprising: a training data generation unit that generates training data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in a time series; a true-value data construction unit that constructs true-value data in which a plurality of true-value vectors indicating the authenticity of each of the plurality of training target images are arranged to correspond to the time series of the plurality of feature vectors; a learning unit that uses the training data and the true-value data to train a three-layered neural network using reservoir computing and acquires coupling constants of the three-layered neural network as a result of the training; an application data generation unit that generates feature vectors as application data from an inference target image, which is an image that is the subject of inference; and a processing unit that uses the acquired coupling constants to generate a trained three-layered neural network and inputs the application data to the trained three-layered neural network, thereby inferring the authenticity of the inference target image.

10. A program causing a computer to function as: a training data generation unit that generates training data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in a time series; a true-value data construction unit that constructs true-value data in which a plurality of true-value vectors indicating the truth or falsity of each of the plurality of training target images are arranged corresponding to the time series of the plurality of feature vectors; and a learning unit that uses the training data and the true-value data to train a three-layered neural network using reservoir computing.

11. A program that causes a computer to function as: an application data generation unit that generates a feature vector as application data from an inference target image, which is an image that is the subject of inference; and a processing unit that uses training data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in chronological order, and true value data in which a plurality of true value vectors indicating the authenticity of each of the plurality of training target images are arranged so as to correspond to the chronological order of the plurality of feature vectors, to train a three-layered neural network using reservoir computing, generates a trained three-layered neural network using the coupling constants of the three-layered neural network obtained as a result of the training, and inputs the application data into the trained three-layered neural network, thereby inferring the authenticity of the inference target image.

12. A learning inference system comprising: a training data generation unit that generates training data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in a time series; a true-value data construction unit that constructs true-value data in which a plurality of true-value vectors indicating the authenticity of each of the plurality of training target images are arranged to correspond to the time series of the plurality of feature vectors; a learning unit that uses the training data and the true-value data to train a three-layered neural network using reservoir computing and obtains coupling constants of the three-layered neural network as a result of the training; an application data generation unit that generates feature vectors as application data from an inference target image, which is an image that is the subject of inference; and a processing unit that uses the obtained coupling constants to generate a trained three-layered neural network and inputs the application data into the trained three-layered neural network to infer the authenticity of the inference target image.

13. A learning method comprising: generating learning data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in a time series; constructing true-value data in which a plurality of true-value vectors indicating the truth or falsity of each of the plurality of training target images are arranged to correspond to the time series of the plurality of feature vectors; and using the training data and the true-value data to train a three-layered neural network using reservoir computing.

14. An inference method comprising: generating a feature vector as application data from an inference target image, which is an image that is the subject of inference; training a three-layered neural network using reservoir computing, using training data in which a plurality of feature vectors corresponding to each of a plurality of training target images are arranged in chronological order, and true-value data in which a plurality of true-value vectors indicating the authenticity of each of the plurality of training target images are arranged so as to correspond to the time series of the plurality of feature vectors; generating a trained three-layered neural network using the coupling constants of the three-layered neural network obtained as a result of the training; and inferring the authenticity of the inference target image by inputting the application data into the trained three-layered neural network.